The semiconductor industry is poised for growth this year and well into the future, and it’s no wonder given continuously high demand for AI computing infrastructure alone.

Global consulting firm Deloitte said semiconductor industry sales are on track to hit a record $697 million this year, and that figure could reach $1 trillion by 2030, which translates into a compound annual growth rate of 7.5 percent for the next several years.

While Nvidia, Intel and AMD command a significant amount of spending for semiconductor products such as CPUs and GPUs, the fast-moving nature of AI development has created an arms race for a much wider field of companies, including startups, to develop groundbreaking silicon solutions that can solve existing bottlenecks and unlock new levels of performance.

There are several semiconductor startups finding momentum in 2025 for this very reason, but their technologies, products and business models vary, ranging from an optical interconnect chiplet that can be integrated into processor designs to "maximize AI infrastructure performance and efficiency," to PCIe-based accelerator cards that are built for high-performance PCs.

Then there are startups solving other important problems, like accelerating big data analytics or enabling silicon-based root of trust for increased security.

What follows are CRN’s 10 hottest semiconductor startups of 2025 so far, which include optical interconnect providers like Ayar Labs and Celestial AI, networking and connectivity vendors like Cornelis Networks and Xsight Labs as well as AI chip designers like Axelera AI and Tenstorrent.

Ayar Labs seeks to speed up large-scale AI workloads in data centers with its high-speed optical interconnect solutions that can eliminate bottlenecks created by traditional methods.

The San Jose, Calif.-based startup in March announced what it called the "industry’s first" optical interconnect chiplet that is compatible with the Universal Chiplet Interconnect Express standard and can "maximize AI infrastructure performance and efficiency while reducing latency and power consumptions." Capable of up to 8 Tbps in bandwidth, the chiplet is intended to integrate in chip designs by other companies.

Ayar Labs made the announcement more than three months after the company said it had raised a $155 million funding round that was led by Advent Global Opportunities and Light Street Capital, with participation from Nvidia, AMD Ventures and Intel Capital.

Axelera AI is fueling AI workloads at the edge with a processor that provides high performance and power efficiency "at the fraction of the cost of alternative solutions." 

The Netherlands-based startup in June launched its global Partner Accelerator Network with more than 15 partners of several types, including OEMs such as Lenovo, Dell Technologies and Advantech; electronics distributors like Astute Group, Rutronic and Silicon Applications Group Corp; and other companies such as Macnica ATD Europe and Arduino.

Axelera AI launched the partner program after announcing in March it was receiving up to 61.6 million in euros from the European Union’s EuroHPC Joint Undertaking to support development of its Titania chiplet for high-performance, energy-efficient and scalable inference. In May, the company said that its Metis AI platform is "fully verified, integrated and configurable through Advantech’s ecosystem" as part of an expanded strategic partnership.

Celestial AI says its optical interconnect technology is uniquely equipped to meet the demands of large-scale AI workloads in data centers "while setting new standards for bandwidth, latency, energy efficiency and total cost of ownership." 

The Santa Clara, Calif.-based startup in March announced that it had raised a $250 million Series C1 funding round led by Fidelity Management and Research Company, with participation from AMD Ventures, Blackrock and Intel CEO Lip-Bu Tan, among several other investors. This brought the startup’s total funding to more than $515 million.

Celestial AI said it will use the new funding to "expand and qualify its volume manufacturing supply chain to serve customer demand" for its products, which consist of connectivity, switching and packaging solutions for optical scale-up networks that are all based around the company’s Photonic Fabric technology platform.

Cornelis Networks, an Intel spin-off, says its family of scale-out networking solutions offer "the best performance with devastatingly good price-performance" for AI data centers compared to offerings based on InfiniBand or Ethernet.

The Wayne, Pa.-based startup in June launched its 400-Gbps CN5000 family of scale-out networking solutions and revealed an updated road map that promises Ethernet compatibility with the next-generation, 800-Gbps CN6000 family in 2026 and Ultra Ethernet compatibility with the 1.6Tbps CN7000 family in 2027.

Cornelis Networks said its Omni-Path chip architecture allows the CN5000 networking products to provide two times higher message rates, 35 percent lower latency and up to 30 percent faster performance for high-performance computing workloads like computational fluid dynamics in comparison to Nvidia’s 400-Gbps InfiniBand NDR solution. The company said the products can also deliver six times faster collective communication for AI applications than RDMA over Converged Ethernet (ROCE) solutions.

EnCharge says its accelerator chip can deliver high performance for AI applications on PCs at a fraction of the cost and power consumption of discrete GPUs.

The Santa Clara, Calif.-based startup launched its Encharge EN100 chip In May, saying that the M.2 version can deliver up to 200 trillion operations per second of AI compute power in an 8.25-watt power envelope. The PCIe version, on the other hand, packs four NPUs to provide roughly 1 quadrillion operations per second, allowing it to provide "GPU-level compute capacity at a fraction of the cost and power consumption." 

EnCharge launched the chip after announcing in February that it had raised a $100 million Series B funding round that was led by Tiger Global, with participation from several other investors. The company has also bolstered its leadership team this year with the hiring of Shwetank Kumar as chief scientist, Jason Huang as vice president of finance and Leslie Szeto as director of human resources.

Lightmatter seeks to unlock major leaps in performance for AI data centers with its 3-D-stacked silicon photonics engine that can connect thousands of processors "at the speed of light." 

Backed by $850 million from investors with a $4.4 billion valuation, the Mountain View, Calif.-based startup in March announced the Passage M1000, which it called the "world’s fastest AI interconnect" for next-generation processors and switches, along with the Passage L200, which it said is the "fastest co-packaged optics" for processor and switch silicon designs.

Lightmatter said the M1000 enables a "record-breaking 114 Tbps of total optical bandwidth" and "connectivity to thousands of GPUs in a single domain." The L200, on the other hand, "enables over 200 Tbps of total I/O bandwidth per chip package, resulting in up to eight times faster training time for advanced AI models." 

Speedata is set to speed up big data analytics workloads "by orders of magnitude" with its "first of-its-kind" Analytics Processing Unit.

The Tel Aviv, Israel-based startup in June announced the launch of its APU, a custom-designed chip called Callisto, alongside a $44 million Series B funding round it had raised from several investors, including Intel CEO Lip-Bu Tan as well as the venture capital firm he founded, Walden Catalyst Ventures, plus Mellanox Technologies co-founder Eyal Waldman.

Speedata said that the Calisto APU, which serves as the basis for the C200 PCIe card for standard servers, was demonstrated to provide a 280-fold performance boost for a pharmaceutical workload over a non-specialized processing unit. This is possible because the APU was designed to "remove the core constrictions that plague modern analytics infrastructure and processing," according to the company.

Tenstorrent is shaking up the AI computing landscape with a business model that combines selling specialized processors with providing open-source software, licensing chip technologies for others to use and working with other firms to develop computing solutions.

The Toronto, Ontario-based startup announced in December that it had raised a Series D funding round of more than $693 million at a pre-money valuation of $2 billion. The round was led by Samsung Securities and AFW Partners, with participation from several other investors, including Bezos Expeditions, Hyundai Motor Group and LG Technology Ventures.

More recently in the past few months, Tenstorrent launched its Blackhole PCIe cards that are "built to handle massive AI workloads efficiently" as an "infinitely scalable solution," and the company entered a strategic partnership with a United Arab Emirates-based startup called AIREV to "co-develop and deploy a high-performance generative AI stack designed for enterprise and sovereign use." 

Xsight Labs is challenging the paradigm for end-to-end connectivity in data centers and edge environments with its high-bandwidth, low-latency DPU, server and Ethernet switch products.

The Kiryat Gat, Israel-based startup announced in May the launch of its Arm-based E1-SoC system-on-chip for cloud and edge AI data centers, saying that it’s the "industry’s highest performance, software-defined DPU" and the "only product of its kind to provide full control plane and data path programmability." 

Xsight Labs said the E1-SoC powers the E1-Server, which it called the "first-to-market 800G DPU." The single rack-unit system can be used as a "high-performance, stand-alone edge server system" or as "comprehensive product development platform," according to the company. It said the E1-SoC is "perfect for developing compute, networking, security, storage, smart switch and air-cooled inference in smaller form factors." 

ZeroRISC says it can secure the "entire supply chain from the factory to in-field operation" with its open-source silicon designs and device management software.

The Boston-based startup in June announced that it had raised an "oversubscribed" seed round of $10 million that was led by Fontinalis Partners, with participation from other investors, including Analog Devices co-founder Ray Stata and SemiAnalysis founder Dylan Patel.

ZeroRISC is based on OpenTitan, the "first open-source silicon root of trust and first-ever open-source chip to reach widespread commercial availability" that was started by the company’s founder and CEO, Dominic Rizzo. The startup said the implementation of its chip designs into products "ensures that security starts below the operating system, offering protection against the most sophisticated hardware and firmware attacks and more common software vulnerabilities.

The 10 Hottest Semiconductor Startups Of 2025 (So Far)

The semiconductor industry is experiencing rapid growth driven by high demand for AI computing infrastructure. The article highlights 10 emerging semiconductor startups in 2025, focusing on innovations in optical interconnects, AI acceleration, networking solutions, and secure silicon technologies. These companies are securing significant funding and partnerships to advance their technologies for data centers, edge computing, and high-performance computing workloads.